Improved verbosity, quality of parser and converter combinations, performance, sorted collections
**PARSER/CONVERTER COMBINATION ENGINE**
* replaced `None` with an explicit `JOKER` for all 'find' parsers/converters methods. Added sanity check at parser registration time: check that parser handles `JOKER` properly
* split `is_able_to_parse` and `is_able_to_parse_detailed` (and the same for convert) so that the first returns a single boolean while the second returns a tuple. This fixed a number of bugs with parser/converter chain generation
* fixed bug with generic parsers by now using the same piece of code than others for registration
* generic parsers and converters are now allowed to chain with other converters. `is_able_to_convert` is therefore now called at runtime on the value generated because it could not fit the required input of the next element
* fixed a few errors in the plugins so that their chaining is correct
**API**
* better error message when the provided location has an extension
* better separation of logging messages between info and debug
* created a true base `ConversionException`
* added package-level methods to access capabilities easily
* `get_capabilities...` methods now do not return some keys when they are empty: the result is more compact to read
* Added `OrderedDict` so that parsing collections now returns a sorted `list/set/dict/tuple`
**PERFORMANCE**
* Big performance improvement for `_get_constructor_signature`, thanks to profiling
* Big performance improvement for `var_checker`, thanks to profiling
* added singleton cached instance for `RootParser` with default plugins so as not to pay the registration costs several times
**TRAVIS INTEGRATION**
* tests, code coverage, doc generation, and releases deployment on PyPI are now automated